>>11518640I'd say half of it is legacy reason and half of it is that there's really not much reason for new people not to use it for numerical work.
I think a lot of physicists aren't actually interested in programming and instead just use what they've used before: For example, one guy at my institute (we're doing computational condensed matter) has been using F77 since the early-to-mid-eighties, and never bothered to learn any of the new features added with F90 (I doubt he even heard of them, actually). He also just keeps reusing code he got from a friend a few decades ago that was written pre-F77.
Other people here heavily use (and develop) computational libraries that are written in Fortran (mostly 90).
From a performance viewpoint, Fortran (compilers) are as fast as the ones for any language. I also think that Fortran's syntax fits pretty well with how physicists would expect things ought to work. Compared to C, working with arrays a lot doesn't require handling pointers directly (until F90, pointers didn't exist at all).
Personally, I stick with Mathematica (because I can get results out of it very quickly) and Fortran (because unlike Mathematica, it doesn't turn every program into unreadable gibberish). I don't really like Python, but many others use it a lot.
>>11518665This isn't true for most BLAS, actually. The reference implementation is written in Fortran, but many actual implementations (like OpenBLAS) use C and a lot of assembler.
